is the adjectival suffix -al a strong suffix?

HAL Id: halshs-01386269https://halshs.archives-ouvertes.fr/halshs-01386269

Submitted on 25 Oct 2016

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.

Is the Adjectival Suffix -al a Strong Suffix?Quentin Dabouis

To cite this version:Quentin Dabouis. Is the Adjectival Suffix -al a Strong Suffix?. Anglophonia / Caliban - FrenchJournal of English Linguistics, Presses universitaires du Midi, 2016, Approches pluridisciplinaires enphonologie anglaise, �10.4000/anglophonia.754�. �halshs-01386269�

https://halshs.archives-ouvertes.fr/halshs-01386269

https://hal.archives-ouvertes.fr

Anglophonia21 (2016)Approches pluridisciplinaires en phonologie anglaise

................................................................................................................................................................................................................................................................................................

Quentin Dabouis

Is the Adjectival Suffix -al a StrongSuffix?................................................................................................................................................................................................................................................................................................

WarningThe contents of this site is subject to the French law on intellectual property and is the exclusive property of thepublisher.The works on this site can be accessed and reproduced on paper or digital media, provided that they are strictly usedfor personal, scientific or educational purposes excluding any commercial exploitation. Reproduction must necessarilymention the editor, the journal name, the author and the document reference.Any other reproduction is strictly forbidden without permission of the publisher, except in cases provided by legislationin force in France.

Revues.org is a platform for journals in the humanities and social sciences run by the CLEO, Centre for open electronicpublishing (CNRS, EHESS, UP, UAPV).

................................................................................................................................................................................................................................................................................................

Electronic referenceQuentin Dabouis, « Is the Adjectival Suffix -al a Strong Suffix? », Anglophonia [Online], 21 | 2016, Online since01 July 2016, connection on 22 September 2016. URL : http://anglophonia.revues.org/754 ; DOI : 10.4000/anglophonia.754

Publisher: Presses universitaires du Mirailhttp://anglophonia.revues.orghttp://www.revues.org

Document available online on:http://anglophonia.revues.org/754Document automatically generated on 22 September 2016.Anglophonia – French Journal of English Linguistics est mis à disposition selon les termes de la licence CreativeCommons Attribution - Pas d'Utilisation Commerciale - Pas de Modification 4.0 International.

http://anglophonia.revues.org

http://www.revues.org/

http://anglophonia.revues.org/754

Is the Adjectival Suffix -al a Strong Suffix? 2

Anglophonia, 21 | 2016

Quentin Dabouis

Is the Adjectival Suffix -al a Strong Suffix?1. Introduction1

1 Outside of the approach introduced by Guierre (1979), few studies on English phonology havesought to give an extensive account of English stress by using quantitative data (Collie 2007,2008; Wenszky 2004). Using a restricted set of a given class of words (e.g. Chomsky & Halle1968: 81, only provide 15 example adjectives in -al to illustrate the rule they propose)2 mightlead to oversimplification or inaccuracy. Without a corpus study, it is impossible to establishthe productivity of a given class, the efficiency of the rule applying to it (if any) and the list ofexceptions to the rule (if any). These are concepts which have been widely used in Guierre’sapproach, whose main characteristics will be presented in §2. Any corpus analysis requiresa theoretical framework in order to interpret the data, and we will adopt Fournier’s (1998,2010) definition of what a strong suffix is for the present study. In § 3, we will present thereasons which have lead us to study -al, a suffix that has been the object of special treatmentsin most extensive works on English stress. Then, we will present how we constituted ourcorpus (§4) and how we analysed it, subset by subset (§5-6). Eventually, the overall resultswill be discussed and compared with the classical weight-based analysis of this suffix’s stressbehaviour (§7).

2. Theoretical Approach2.1 The Guierrian School

2 The “Guierrian School” is an approach which was introduced in the seventies by Guierre(1979). Its two main characteristics are the use of pronouncing dictionaries3 to study thephonology of English and the use of orthography when necessary (e.g. elements such asconsonant geminates, vocalic digraphs or final mute <e> are taken into account).4

3 Within that approach, only three levels of stress are acknowledged:• primary stress (annotated [ˈ] or /1/)• secondary stress ([ˌ] or /2/)• no stress (/0/)

4 English stress is also described as being regulated by the following four general stressprinciples (Fournier 2007, 2010):

1. Every lexical unit5 has one and only one primary stress.2. There can be no sequence of two stresses within a lexical unit.3. No lexical unit can begin with two unstressed syllables.4. Syllables which receive neither stress /1/ nor stress /2/ are unstressed.

5 In this paper, we will make a distinction between derived and non-derived entries, andtherefore need to specify what we mean by derivation. Following Fournier (2010: 39), we willconsider that the “base is the closest form attested in the English vocabulary”. This means that,even though the base and its corresponding derivative are usually connected semantically, itmay not be the case (e.g. universe → university). In the rest of the paper, “derived entries”will refer to suffixed units formed by suffixation to a free base. In the case of suffixation, twotypes of derivation are acknowledged:

• juxtaposition (e.g. origin → original; mayor → mayoral; nature → natural)• substitution (e.g. colossus → colossal; inferno → infernal; puberty → pubertal)

6 Suffixes can be attached to free bases (e.g. continent → continental) or bound bases (e.g.infern- + -al). Therefore, words like mortal are seen as being non-derived but suffixed. Suchcases raise the question of how far suffix recognition should go. How can we distinguish asuffix from an ending that just happens to look like a suffix? We adopt the view that suffixes

are categorisers (Aronoff 1976: 71; Marchand 1969) and therefore that it is the association ofa specific sequence of phonemes (or graphemes) with a specific category6 which allows us toidentify whether we are dealing with a suffix or an ending.

2.2 Vowel Qualities and Values7 In the analyses below, we will use the terminology displayed in Table 1.

Table 1. Correspondences between orthography and pronunciation for stressed vowels (afterFournier, 2010: 98)7

8 The table shows that each orthographic vowel can have:• four8 different values when it is a monograph (e.g. r-coloured, checked, free and r-

coloured free);• two different values when it is a digraph (e.g. free and r-coloured free).

9 The possible realisations of a given orthographic vowel (e.g. [ɑː], [ӕ], [eɪ], [eә] for <a>) aresaid to share a quality. The value of a given vowel is conditioned by the context in whichit appears, the detailed rules are given in Fournier (2010).9 A reviewer points out that theterms “checked” and “free” are often used to describe syllable structure. This is not whatFournier’s (2010) system does, even though these different values are often correlated withsyllable structure. In his system, values correspond to a specific relationship between spellingand pronunciation, e.g. V:̆ <a> - [æ], <e> - [e]; V:̅ <a> - [eɪ], <e> - [iː]…

10 Crucially, the Guierrian School sees vowel values as being predictable from the contextsurrounding vowels (mainly the right-hand consonantal context). Therefore, vowel values arephenomena that the theory seeks to predict along with the position of stresses. This meansthat stress patterns cannot be argued to depend on the presence of long vowels, contrary towhat is proposed in most generative approaches which see vowel values as lexical (see §7 forfurther discussion).

2.3 Strong Suffixes and Strong Endings2.3.1 Strong Suffixes vs. Neutral Suffixes

11 In the literature on English stress, it is commonly admitted that English has two types ofsuffixes. The names, descriptions and lists of these suffix classes vary considerably betweentheories, but there are constants.

12 Some suffixes affect the position of stress, and have other properties which differ fromsuffixes which do not affect stress. After Fournier (2007), we shall call these suffixes “strongsuffixes”, but they are called “stress-imposing suffixes” in pronouncing dictionaries, “stress-placing suffixes” (Burzio 1994), “Class I suffixes” (Siegel 1974), “Level I suffixes” or “stem-level suffixes” in Lexical Phonology (Kaisse & Shaw 1985; Kiparsky 1982) and the theoriespreserving the idea of stratification like Stratal OT (Bermúdez-Otero & McMahon 2006;Bermúdez-Otero 2012). In generative phonology, this distinction was based on the distinction



made in early works like Chomsky & Halle (1968) between suffixes associated with the “+”boundary (stress-affecting suffixes) and those associated with the “#” boundary (stress-neutralsuffixes). Strong suffixes can be attached to bound roots and free roots alike and they tend to bevowel-initial whereas stress-neutral suffixes attach only to free roots and tend to be consonant-initial.

13 The formal description of what these suffixes “do” and the theoretical apparatuses whichhave been proposed vary considerably. That variation can partly be explained by the rangeof phenomena that a theory is trying to account for and many phenomena can be linked withaffix classes (e.g. alternations, allophony, semantic compositionality, affix ordering; see Benua(1997: §5.2.2) and Giegerich (1999: Ch. 2) for discussion of some of these phenomena). TheGuierrian School mainly seeks to account for two phenomena: the placement of stresses andthe values taken by stressed10 vowels. Let us consider two definitions, from two Guierrianauthors, which show where part of the problem lies.

Some final sequences apply an absolute constraint by fixing primary stress immutably on a givensyllable” (Trevian 2003: 55).

Strong endings impose direct computation, i.e. the relevance of the unit’s structure and only thatstructure” ( Fournier 1998).11

14 Based on the common observation that words derived with these strong suffixes shareproperties with non-derived words (e.g. Luick’s rule12 applies to them), it has been proposedthat the rules regulating stress in both derivatives with strong suffixes and non-derivedwords take place at the same “level” or in the same phonological domain, either because ofidiosyncratic properties of suffixes or because of their segmental structure (Raffelsiefen 2015).Some suffixes are associated with stress patterns which are not typical of simplex words (e.g.“French” auto-stressed suffixes like -eur/euse, -aire, -ette, -esque, or words in -ic13, whichare always stressed on their penultimate syllable), whereas other words with different strongsuffixes have stress patterns which are comparable to those of simplex words (e.g. adjectivalsuffixes associated with penultimate stress when preceded by a consonant cluster: attentive,elemental ≈ adventure, coriander, antepenultimate stress for a number of suffixes such as -ity, -ify, or -(at)ion: acidify, vanity ≈ ambassador, seminar). Strong suffixes with non-typicalstress patterns should most likely be accounted for in terms of idiosyncratic suprasegmentalproperties of suffixes, which is compatible with Trevian’s (2003) definition or Selkirk’s (1980)proposal that some suffixes have a lexically stored stress foot structure. However, this analysiscannot be extended to strong suffixes which are associated with stress patterns that are typicalof simplex words.

15 Fournier (1998) analyses all strong suffixes as sharing the property of triggering what hecalls “direct computation”, i.e. the fact that the placement of primary stress for derivativesis not influenced by the position of stresses in their base. In contrast, neutral suffixes trigger“computation by reference”, i.e. stress preservation from the base. Moreover, he argues thatwhen the resulting stress pattern is the same as that of simplex words with a comparablestructure, it is not possible to demonstrate that strong suffixes do anything but trigger adirect computation of stress. This contradicts Trevian’s (2003) view that they would fix stresson a given syllable, which seems to be based on the observation of a correlation, not on ademonstration. The two definitions differ when analysing a case such as proˈcedure [iː]→proˈcedural [iː]. Even if the derivative’s stress pattern conforms to the general proparoxytonepattern proposed by Trevian, his model does not account for the difference between this caseand a case for which Luick’s rule applies (e.g. ˈnature [eɪ] → ˈnatural [æ]). Therefore, thedefinition of a strong suffix which is adopted here will be Fournier’s (1998, 2010).

2.3.2 Suffixes and Endings16 We can only demonstrate that direct computation applies if there is a suffixation, but it does

not mean that the trigger is necessarily suffixation. As Guierre (1979) points out, certain“endings” constitute such triggers. For example, some adjectival suffixes are associated witha penultimate stress pattern when they are preceded by a consonant cluster and may not haveany effect on pronunciation when this is not the case. Additionally, and this is of particular



relevance to our study of -al, some suffixes trigger direct computation when they are precededby a certain suffix (e.g. -ival, -inal, -iacal, -inous, -ulous). Therefore, direct computation istriggered by the association of two elements (the suffix and either a consonant cluster oranother suffix) and not one alone. Henceforth, these structures have been referred to as “strongendings” (Fournier 2007; Guierre 1975).

17 A structure such as -{i, e, u}+V(C0(e)), which is regularly associated with a proparoxytonestress pattern can be made of suffixal and/or non-suffixal material (e.g. endings in -ious,-ual or -eal in which the first vowel can be analysed as being part of the root or as anextender, see Bauer et al. 2013: 181). In other words, suffixation creates the environment whichtriggers direct computation but it does not mean that suffixation alone causes that phonologicalbehaviour. It can either be the presence of both a certain suffix and certain elements precedingit (e.g. a consonant cluster, a vowel or another suffix) or simply the fact that the addition ofthe suffix creates a certain sequence that triggers direct computation.14

3. Why -al?18 This paper is far from being the first dealing with this suffix, but the question which we will

try to answer is one that has not often been asked. In the literature, we can find discussionsabout -al’s combinatorial properties, especially when it follows -ment (Aronoff 1976: 53-56;Giegerich 1999: 78-81; Hay 2002). We can also find the mention of -al among a specialsubclass of suffixes which are associated with not one but two stress patterns. Wells (2008)describes this suffix as “impos[ing] stress one or two syllables back”. We can most oftenfind the stress patterns of adjectives in -al described as being regulated by the following twogeneralisations (Burzio 1994: 202; Chomsky & Halle 1968: 81; Marchand 1968: 244)15:

• if it is preceded by a heavy syllable, the word is stressed on that syllable.• if it is preceded by a light syllable, it is stressed on the antepenultimate syllable.

19 The main goal of this paper is to analyse -al within the framework of the Guierrian School butthis weight-based analysis will be discussed in §7. Let us now consider the Guierrian analysesof this suffix.

20 Trevian’s (2003: 113) analysis is close to the classical weight-based analysis. He divides thefirst rule into two: words for which -al is preceded by a consonant cluster (e.g. eˈternal)or whose base has a long vowel in their final syllable (e.g. ˈanecd[əʊ]te → ˌanecˈdotal) arestressed on their penult. He adds a rule for words ending in -val, which also have penultimatestress (e.g. adjectival, medieval).16

21 Fournier (2007, 2010) does not give a comprehensive analysis of this suffix, but proposes thefollowing rules17:

• -C2 + adjectival suffix → /(-)10/• -oidal, -ival → /(-)10/• -ical, -acal, -inal → /(-)100/

22 Words which are not subjected to these rules are implicitly treated by more general rules whichare not specific to -al (e.g. {i, e, u}+V(C0(e)) → /(-)100/, neoclassical compounds, neutralderivation, Normal Stress Rule). This means that he probably does not consider -al to be astrong suffix in itself, but only in these specific environments. Therefore, our study seeks tocorroborate this intuition empirically.

23 We can see that the analyses of -al, even if they have some common points, divergeconsiderably. It seems that Fournier’s analysis is the most detailed, even though some partsof his analysis need to be made explicit. This paper will attempt to do so and to answer thefollowing questions:

• How can we describe the phonological effects of -al? In other words, what can beattributed to -al itself, to combinations of parameters (e.g. presence of the suffix andnumber of preceding consonants) or to wider generalisations?

• Is it accurate to describe the stress pattern of -al adjectives as being determined by theweight of the preceding syllable?

4. Corpus

4.1 Constitution24 In the tradition initiated by Guierre (1979), we contend that the best way to analyse and

understand the behaviour of this suffix is a corpus study. The different steps which we followedto constitute this corpus are described below, and they attest to our will to compile a corpusthat was as complete, representative and reliable as possible.

25 We started with a search in the online Oxford English Dictionary (OED) for non-obsoleteadjectives ending in <-al> in order to be able to access semantic and historical information forall entries in later steps of the study. This search gave 5729 results.

26 Then, we kept only the entries which were present in at least one of the two pronouncingdictionaries that we decided to use for the study: the Cambridge English PronouncingDictionary (CEPD), 17th edition, and the Longman Pronunciation Dictionary, 3rd edition(LPD).18 At that point, the corpus contained 1563 entries.

27 This paper does not aim to study rare or specialised words, which led us to use a frequencythreshold using the Corpus of Contemporary American English (COCAE): we adopted athreshold of one occurrence in the COCAE. The reason for choosing this threshold (which isarbitrary and therefore debatable) is that if an entry has at least one occurrence (out of the over400 million words contained in the COCAE), we can reasonably assume that it is used at leastto some extent, whereas when no occurrence can be found in the COCAE, it is impossible toknow whether the entry is rarely used or not used at all in contemporary English.19 This choiceled to the exclusion of 295 entries.

28 In order to avoid redundancy in our corpus, we decided to exclude two types of semanticallytransparent tokens:

• The ones whose first element is a prefix (e.g. bi-, in-, pre-,...)• The ones whose first element is a neoclassical root (e.g. bio-, equi-, psycho-,...)

29 In these constructions, the second element (e.g. natal in prenatal) systematically preservesits pronunciation, which is why we left them out. Therefore, as long as its second element isalready present in the corpus, the construction was excluded (196 entries). We kept only sixcases for which the second element was not attested in our corpus at that point.

30 Eventually, our corpus contains a total of 1073 entries, out of which 938 are to be found onlyin LPD and 963 only in EPD, 828 entries being common to both dictionaries.

4.2 Organisation for Analysis31 As mentioned above, our aim is to apply an explicit version of Fournier’s (2007, 2010)

analysis of -al. Therefore, we have preserved the rules that he has proposed and have split theremaining words into categories which are not specific to -al. Additionally, we chose to analysedisyllables separately. This is because their structure does not allow for the same method ofanalysis as for longer words. The division of the corpus is shown in Figure 1.20

Figure 1. Division of the Corpus



5. Analysis of disyllabic words32 Following Trevian (2003: 140), we decided to analyse disyllables separately because they are

systematically stressed on their first syllable and because the methods of analysis requiredfor longer words do not apply to dissyllables. Crucially, because of their length, they cannotbe potential candidates for Luick’s rule and there are no possibilities for stress to move withderivation.

33 This set contains 121 entries, to which we added three of the six semantically separableconstructions that we preserved because the second element of the construction was absentfrom the corpus: bipedal, extramural, intramural. For these entries, the relevant part foranalysis here is only this second element, i.e. pedal and mural.

34 51 of the entries are non-derived (i.e. they are composed of a bound root and -al) and all butone (baˈnal) are stressed on their first syllable. Fournier (2010: 136-137) notes that disyllableswith the structure C0VCVC1 do not follow any global generalisation with regards to the valueof their stressed vowel21, i.e. there is a similar proportion of words with a free vowel (e.g. agent,bacon, equal, icon, moment) and words with a checked vowel (e.g. atom, camel, damage,foreign, lizard). However, this is not what we can see in our corpus: the 21 words with thatstructure are all pronounced with a free vowel, with the exception of pedal which has a variantin [e]:

(1) chiral, feral, final, floral, legal, lethal, local, moral, natal, oral, oval, penal, plagal, primal,regal, renal, total, venal, vital, vocal ©

22

35 The remaining 73 entries are all derived, and all are stressed on their first syllable. They allpreserve the pronunciation of their base when derived (apart from a few particular exceptionswhich will be discussed below).

36 We have three cases for which the orthographic vowel is modified (as well as the phoneticone), most likely because the base and the corresponding derivative have a different historicorigin: feast → festal; nose → nasal; pope → papal. Additionally, the case of cloister →claustral could be described in a similar fashion, except that a “regular” derivative can befound in our corpus: cloistral.23

37 In sum, disyllables constitute a fairly regular set, both in terms of stress (one exception withfinal stress) and in matters of pronunciation (systematic free vowels where non-regulationcould be expected).

6. Analysis of Words of Three Syllables or More38 For longer words, we chose to divide the corpus into four subsets: words that have one

of Fournier’s (2010) strong endings (§6.1), neoclassical compounds (§6.2), derived (§6.3)and non-derived (§6.4) entries that do not belong to the first two sets. We expect the firstset to mainly follow Fournier’s (2010) generalisations. Neoclassical compounds are treatedseparately as they are known to have distinct phonological properties (Bauer et al. 2013: 18;Fournier 2010: 76-77; Fournier 2011; Guierre 1979: 737; Tournier 1985: 92). Most suffixes arestrong when attached to these compounds so we expect to find cases of direct computation. Asanalyses diverge on the remaining two sets, we will start by reviewing the empirical facts first.

6.1 Fournier’s Strong Endings39 The method of analysis used in this part of the study is inspired by that developed by Fournier

(1998) which follows from his definition of strong endings. This method consists in theanalysis of preservation or non-preservation relationships between bases and their derivatives.In certain cases, the direct computation of stress and vowel values predict different results thancomputation by reference. These cases will be called relevant cases, as they are the only oneswhich allow us to determine whether a certain suffix or ending is strong. Cases for which thetwo modes of computation make identical predictions will be called non-relevant cases, asthey do not provide any information which may be used to determine the nature of the suffixor ending’s behaviour. In these cases, the stress pattern and the vowel values of derivativesmay be predicted by preservation or direct computation, as in the examples in (2).



(2) plaˈcenta → plaˈcental; eˈlector → eˈlectoral, ˈmargin → ˈmarginal

40 Relevant cases may behave in two ways. If it is direct computation that actually applies, thenstress shifts (see (3)) or vowel shifts (see (4)) occur.

(3) ˈorigin → oˈriginal; ˈmaniac → maˈniacal; ˈindustry → inˈdustrial

(4) cr[aɪ]me → → ˈcr[ɪ]minal; ˈr[aɪ]te → ˈr[ɪ]tual; liˈb[iː]do → liˈb[ɪ]dinal

41 However, if computation by reference applies, then no shifts occur. This may result in pre-antepenultimate stress (see (5)) or antepenultimate stressed vowels that are free24 (see (6)).

(5) ˈspirit → ˈspiritual; beˈhaviour → beˈhavioural; ˌaspiˈration → ˌaspiˈrational25

(6) proˈcedure [iː] → proˈcedural [iː]; ˈf[eɪ]tal → ˈf[eɪ]talist

42 If direct computation applies in the vast majority of relevant cases for a given suffix or ending,then we can conclude that this suffix or ending is strong. However, if computation by referenceprevails, then we can conclude that this suffix of ending is neutral.

43 We applied this method of analysis to the sets of strong endings proposed by Fournier (2010),and the results of this analysis are shown in Table 2.

44 Table 2. Results of the analysis of relevant cases for strong endingsRelevant cases

Stress patternStress shift Vowel shift No shift

Variation /1/ Exceptions Ruleefficiency

-C2+-al 52 + 3 var 0 2 4 2 89/91 (98%)

-oidal 8 0 1 0 1 8/9 (89%)

-ival

/(-)10/

6 0 0 1 0 8/8 (100%)

-ical 64 + 1 var 1 + 2 var 0 2 0 293/293(100%)

-iacal 8 0 0 0 0 9/9 (100%)

-inal 7 + 1 var 3 + 1 var 0 5 0 28/28(100%)

-{i, e, u} +-al

/(-)100/

116 + 7 var 1 1 8 8 244/252(97%)

45 This table presents the numbers of relevant cases per ending. The numbers under the columns“Stress shift” and “Vowel Shift” refer to cases like those in (3) and (4), respectively. Thecolumn “No shift” contains both cases like (5) and (6). The column “Variation /1/” givesthe number of words which have more than one stress pattern and it includes the cases ofvariation already mentioned in the “Relevant cases” columns. Likewise, the “Exceptions”column regroups the exceptions already counted in the “No shift” column along with otherexceptions not included in the relevant cases. Finally, the “Rule efficiency” column gives theratio of words with at least one regular variant on the total number of words with a givenending.

46 Table 2 shows that a description of these endings as being strong endings seems to be accurate.Let us detail the results for the ending -C2+-al to explain why this is so. This set contains:

47 57 relevant cases, among which:• 52 cases of shifts of primary stress to the penultimate syllable:

(7) ˈautumn → auˈtumnal; ˈdocument → ˌdocuˈmental; ˈphantasm → phanˈtasmal,...

• 3 cases which also present an isomorphic variant:

(8) ˈdiphthong → diphthongal /(2)10/ ~ /100/; ˈmatriarch → matriarchal /2010/ ~ /1000/; ˈprefix(n) → prefixal /(2)10/ ~/100/ ©

• no cases of vowel value modification (none are expected as vowels which are followedby C2 are always checked);

• 2 cases where computation by reference applies:

(9) ˈpuberty → ˈpubertal; ˈsinister → ˈsinistral ©

• one additional case of variation: magistral /100/ ~ /010/



• 34 non-relevant cases.26

48 Therefore, there are 55 (if we exclude variants) out of 57 relevant cases which do follow themechanism of direct computation.

49 Moreover, we can see that throughout the whole set, 85 entries are strictly stressed on thepenultimate (i.e. without variants). Thus, the generalisation -C2 + -al → /-10/ can be formulatedand its efficiency is of 93%. This efficiency jumps to 98% if we count all the words whichhave at least a regular variant.

50 All data sets in Table 2 present similar results and can thus be considered to contain strongendings.27

51 As adjectives in -al may have penultimate or antepenultimate stress and that the rest of thecorpus may not be divided into somehow uniform sets, we cannot use exactly the same kind ofevidence. Let us see why by looking at the example of the words in -C2 + -al. If there was nogeneral tendency towards penultimate stress in this set, how could we know whether ˈpuberty→ ˈpubertal constitutes a relevant case in which computation by reference applies or a non-relevant case? We could not. Therefore, in the rest of the analysis, positive evidence for directcomputation will only come from stress or vowel shifts and positive evidence for computationby reference will only come from cases of pre-antepenultimate stress (which is never foundwith strong endings) or antepenultimate stressed long vowels.

6.2 Neoclassical Compounds52 The structure of these compounds can be described as being ABC Guierre (1975: §8):

(A) (B) (C)

root or prefix + root + suffix

Example: fratri- -cid- -al

53 This set contains 55 entries, among which 38 are derived from a free stem and may potentiallyexhibit stress or vowel shifts. Such shifts do occur:

• Stress shifts to a penultimate stress pattern:

(10) ˈgenocide → ˌgenoˈcidal; ˈmenopause → ˌmenoˈpausal; ˈribosome → ˌriboˈsomal...(18 cases)

• Stress shifts to an antepenultimate stress pattern:

(11) ˈhexagon → heˈxagonal; ˈocciput → ocˈcipital; ˈarchitecture → ˌarchiˈtectural... (10 cases)

• One vowel shift: trigon [aɪ] → trigonal [ɪ].54 These shifts are typical of strong endings, but the patterns observed are not uniform. If we

consider the whole set, both derived and non-derived, the following stress patterns can beobserved:Table 3. Stress patterns of neoclassical compounds

/-10/ /(-)100/ /-10/ ~ /(-)100/ /-10/ ~ /(-)1000/ Total

Number ofentries

29 21 3 2 55

55 The two main stress patterns, penultimate and antepenultimate, are found in similarproportions, and it could be assumed that an additional parameter is responsible for thisdistribution. In a similar study on words in -ous, Fournier (2011) found that the root B seemedto play a role in stress assignment in these compounds. He describes the roots that retain stressas “attractive” (penultimate stress), as opposed to “non-attractive” roots which let the NSRapply (antepenultimate stress).

56 Therefore, this parameter was tested and we found it to be relevant for -al as well. That canbe evidenced by two things:

• Uniformity of stress patterns for compounds with the same root B (e.g. allcompounds in -cidal are stressed /-10/; all those in -gonal are stressed /(-)100/);

• Commutation of the root B: following the principle of minimal pairs, Fournier (1997:60) states that “if, in a defined class, a feature divides this class into two groups of unitshaving a distinct behaviour, and that this feature is the only difference between thesetwo groups, then it is necessarily this feature which is responsible for the difference inbehaviour”.28 In our case, if we can find pairs which are distinguished only by their rootB and which present a different stress pattern (see (12)), then we can assume that it isthis root B which determines this difference.

(12) epiˈdural ~ eˈpiscopal ; hexaˈhedral ~ heˈxagonal ; antiˈdotal ~ anˈtiphonal

57 Guierre (1979) suggested that some orthographic factors may account for the stress behaviourof roots. He proposed that those which contain a vowel digraph or end in <Ch> or <Cr> shouldbe attractive. However, as Fournier (2011: 312) points out, these elements cannot account forthe behaviour of all roots. Therefore, the attractiveness or non-attractiveness of a root shouldprobably be attributed to some form of lexical marking.

58 If we follow Fournier’s terminology, we have the distribution shown in Table 4.Table 4. Distribution of roots B in neoclassical compounds

Non-attractive roots Attractive roots Semi-attractive roots29

-gon-, -metr-, -op-, -pher-, -phon-, -pod-, -scop--cipit-, -cultur-, -tectur-, -genit-, -enter-

-cid-30, -drom-, -dur-, -(a)ev-, -hedr-,-paus-, -som-, -typ-, -vir-,

-dot-, -fug-, -pet-, -voc-

59 Table 4 shows that Guierre’s (1979) orthographic criteria for predicting the stress behaviourare not really satisfactory. The only two roots with a vowel digraph are indeed attractiveroots but two roots end in <Cr>: one is attractive (-hedr-) and the other is non-attractive (-metr-). Interestingly, all disyllabic roots are non-attractive. Clearly, more research is neededto establish whether there are formal reasons why a root is attractive or if it is mostly lexicalmarking.

60 Overall, this set shows that the presence of -al alone does not allow us to predict the observedpatterns. To do so, we need to take into account both the presence of the suffix and that ofa certain type of root B.

6.3 Remaining Derived Entries61 This set contains all the derived entries which do not belong to any of Fournier’s (2010)

strong endings (which were treated in §6.1) and that are not neoclassical compounds (whichwere treated in §6.2). It contains 164 entries, including 102 entries in -ional. These entriesstrictly preserve the pronunciation of the base from which they are derived, with the oneexception of ˈnation [eɪ] → ˈnational [æ], which also happens to be the only entry for whichthe sequence -at- belongs to the root, i.e. nat- (as in native, innate). In this case, the differencein morphological structure is reflected in the phonological behaviour of the derivative.

62 To interpret these results, one needs to determine whether -ion should be considered as a singlephonological syllable or as two. Consider the alternations in (13).

(13) ocean [ˈəʊʃən] à oceanic [ˌəʊʃiˈænɪk]partial [ˈpɑːʃəl] à partiality [ˌpɑːʃiˈæləti]

63 It can be argued that derivation “reveals” that two phonological vowels are compressed intoone under reduction and synaeresis. However, -ion is very rarely suffixed with strong suffixeswhich would place stress on one of its two syllables so that these syllables could be revealed. Asearch in the OED gave us only two such cases: regionic [ˌriːdʒɪˈɒnɪk] and visionic [vɪʒɪˈɒnɪk].31

These cases seem to show the phonological dissyllabic nature of -ion, but they are so marginal,and unattested in our pronouncing dictionaries, that there seems to be very few arguments fora dissyllabic interpretation of -ion apart from the fact that it belongs, orthographically, to thelarger class of words in -{i, e, u}+V(C0(e)).

64 To test if a generalisation such as “-al → /(-)100/” applies in this set, we need to consider twooptions: either -ion is monosyllabic or it is disyllabic. We will not attempt to solve the issue

of the number of syllables in -ion here but we will simply consider how the two options leadto different interpretations.

65 If -ion is monosyllabic, then stress is antepenultimate in all the entries considered. Therefore,the only way to determine if -al is affecting their pronunciation is by considering vowel values.As stress is antepenultimate, we could expect Luick’s rule to apply. However, this is not thecase; free vowels in bases are systematically preserved, with the exception of national. Thereare 40 bases containing a V ̅≠ 32 which is preserved in their corresponding derivatives.These being the only cases for which the data can be interpreted (bases with V ̆being non-relevant), it would seem that -al is neutral in that set.

66 If -ion is disyllabic, then stress is ante-antepenultimate and -al is strictly neutral in the wholeset.

67 Therefore, both interpretations lead to the same conclusion: -al is neutral after -ion.68 After removing the 102 entries in -ional, we are left with 62 entries in this set. If we follow

Fournier’s (2010) implicit analysis of -al, we would expect it to be neutral in this set. However,the situation is more complex.

69 There are 42 cases for which stress does not shift, let us consider the 5 cases in (14).

(14) procedure [iː] → procedural [iː] femur [iː] → femoral [e]nature [eɪ] → natural [æ]ephemera [e]/[iː] → ephemeral [e]/[iː]zenith [e]/[iː] → zenithal [e]/[iː]

70 The first case indicates the neutrality of -al, as the vowel is not shortened to [e], contraryto Luick’s rule. Now, incidentally, the following two cases behave in the opposite way, andLuick’s rule seems to be applying. Finally, the last two cases present variation in both the baseand the derivative when we could expect only a V ̆if Luick’s rule was applying.

71 There are also cases where the fact that the position of stress does not shift seems to indicatethe neutrality of -al if a rule -VCal → /(-)100/ was applying, as those in (15).

(15) ˌarchˈduke → ˌarchˈducal; auˈrora → auˈroral; cloˈaca → cloˈacal; deˈcree → deˈcretal;beˈhaviour → beˈhavioural

72 If we now consider words for which stress does shift, the 20 entries can be divided into threegroups33:

(16) shift to penultimate stress, with no variants: anecdotal, azimuthal, hormonal, paradisal,patronal34 ©

(17) shift to antepenultimate stress, with an isomorphic variant with penultimate stress: artisanal,coronal or with pre-antepenultimate stress: prefectural, or both isomorphic and penultimatevariants: pyramidal ©

(18) no shift, with a variant in which stress shifts to the penultimate syllable: apsidal, cantonal,cerebral, communal, doctoral, epochal, electoral, integral, mayoral, palatal, skeletal ©

73 These 62 entries are the only ones in which -al, by itself, could be held responsible fortriggering direct computation if it were to occur. In the case of Fournier’s (2010) strong endingsor neoclassical compounds, we saw that stress is determined by the suffix and certain elementspreceding it. In this last set, there is no common preceding element to be found. We do findsome evidence for direct computation but it does not seem to be the general case in this set.Moreover, the only set which could really indicate what -al “does” by itself appears to be themost unstable set of all.

74 As suggested by a reviewer, the heterogeneity of the stress patterns of -al adjectives mightbe connected with the lack of productivity of that suffix. The lack of an active pattern couldexplain the lack of uniformity. The OED seems to confirm that -al is indeed unproductive,as it lists only 251 new adjectives between 1950 and 1999 and these are mostly specialisedwords (e.g. ambilineal, audiological, contrafactual). It also lists one new adjective between2000 and 2016, cissexual, which a case of prefixation.



6.4 Remaining Non-Derived Entries75 This set contains 34 entries, and we will also mention the 6 labelled “Other” in Figure 1, which

are either compounds or opaque prefixed constructions.76 Out of the 34 entries, 32 follow the NSR, i.e. are stressed on their antepenultimate syllable.

The two entries in (17) present variation with a penultimate stress pattern.

(19) decanal /010/ ~ /100/palpebral /100/ ~ /010/

77 The set is fairly stable, and these entries are stressed just as regular non-derived items usuallyare.

78 We separated two minor classes: opaque prefixed constructions and compounds, as thesestructures usually entail specific stress patterns: the former may not receive primary stresson their prefix unless they are nouns (i.e. adjectives, verbs, prepositions, adverbs) whereascompounds tend to be stressed on their first element (Fournier 2007).

79 The corpus contains 4 inseparable prefixed constructions, and 3 of them do not violate therule: ocˈclusal, iˈnaugural, reˈciprocal. However, only the first one could present a potentialconflict with the NSR, thus it seems to be the only relevant case showing that the rule doesapply. The fourth entry does not follow the rule: ˈconjugal. Due to the small number of entriesand of relevant cases of conflict with the NSR, it seems difficult to draw any conclusionsregarding the rule of opaque prefixed constructions other than nouns.

80 Eventually, the corpus contains two compounds: ˌteeˈtotal and ˌviceˈregal, the stress patternof which is not that of regular compounds (which are usually stressed on their first element).However, we will refrain from drawing any conclusions from just two compounds.

7. Discussion81 We started with two questions regarding -al, let us now consider what answers the results

provide.82 How can we describe the phonological effects of -al? In other words, what can be attributed to

-al itself, to combinations of parameters (e.g. presence of the suffix and number of precedingconsonants) or to wider generalisations?

83 There are two sets for which it is not possible to determine the effects of -al because of a“conspiracy” of rules: non-derived entries and -{i, e, u}+-al. In the absence of evidence toshow that -al is indeed responsible for the stress patterns observed, we can assume that moregeneral rules, respectively the NSR and the rule -{i, e, u}+V(C0(e)) → /(-)100/, apply.

84 We have also seen a number of sets where we found evidence of direct computation, but thisseems to be the case only when -al is associated with an element preceding it. This elementcan be:

• a consonant cluster;• another suffix in -ival, -oidal, -ical, -iacal and -inal;• a root B in neoclassical compounds.

85 In these cases, -al is only one of the components of strong endings triggering directcomputation, even if it is -al suffixation which makes the demonstration possible. What isstriking here is the great regularity of “micro-paradigms” (Girard, 2007) such as -ival or -oidal(both comprise under 10 entries), even though their stress pattern (penultimate) is “marked” forthis structure. Moreover, if there was a rule -VCal → /(-)100/, these micro-paradigms wouldbe exceptional.

86 We saw that -al was neutral after -ion. Can this be an argument to say that -al, by itself, is nota strong suffix? Two answers may be proposed:

• Yes: words in -iac and in -ion both have the structure -{i, e, u}+VC0, which is part ofthe larger rule -{i, e, u}+V(C0(e)) → /(-)100/. In general, the last strong suffix or endingtriggers direct computation, which means that -al, if it were strong, should trigger directcomputation in both environments. However, this is not the case for -ional, which maysuggest that -iacal is an exceptional micro-paradigm and that -al is stress-neutral after all.

• No: Fournier (2010: 34) points out the fact that some words with the structure -{i,e, u}+VC0 neutralise -ate, which is strong in other environments, as exemplified byˈalienate, afˈfectionate, comˈpassionate, ˈmatriarchate, ˈpassionate,... Outside of thisenvironment, -ate regularly shifts stress to the antepenultimate, e.g. ̍ origin → oˈriginate,saˈliva → ˈsalivate. Therefore, it could be proposed that -{i, e, u}+VC0 is neutralisingfor (at least certain) suffixes attached to it and that -al may be strong but that it is simplyneutralised here, with the exception of the micro-paradigm -iacal.

87 Since there is no principle way to choose one of the proposed answers, this set will beconsidered irrelevant to determine whether -al is strong or not. However, in both cases, -iacalshould be analysed as exceptional.

88 Only the set discussed in §6.3 is left to determine -al’s status and it is that of the remainingderived entries. As we have seen, this set is the least regular of all. Therefore, we cannot saythat -al is strong in that set, but we cannot say it is completely neutral either.

89 To sum up, the generalisations on Fournier’s (2010) strong endings and the NSR do wellwhen confronted with the data. Neoclassical compounds constitute a special set, whose stresspatterns are constrained by the interaction between the root B and -al. However, the data donot allow a larger generalisation of the type -VCal → /(-)100/.

Is it accurate to describe the stress pattern of -al adjectives as being determined by the weightof the preceding syllable?

90 In this section, we will consider the data discussed in this paper under a different anglethan that which has been adopted so far. We cannot answer our second question withoutspecifying whether we are trying to account for surface patterns or to predict stress patternsfrom underlying vowels.

91 The weight-based account of the stress patterns of -al adjectives works fairly well if we areonly looking at surface patterns. Virtually all words in the corpus follow the generalisationstating that words with heavy penults have penultimate stress and that words with light penultshave antepenultimate stress.35 We only found the following 11 exceptions:

(20) words with closed penults that do not have /(-)10/: dipthongal /010/ ~ /100/, matriarchal/2010/ ~ /1000/, prefixal /100/ ~ /010/, sinistral /100/ and magistral /100/ ~ /010/;

(21) words with /(-)10/ and ˈV ̆ in open penults: ˌaziˈm[ʌ]thal, ˌartiˈs[æ]nal (mainpronunciation /0100/), ˌpyram[ɪ]dal (main pronunciation /0100/ + variant in /1000/), canˈt[ɒ]nal(+ variant in [əʊ] + variant in /100/), eˈp[ɒ]chal (main pronunciation /100/), inˈt[e]gral (mainpronunciation /100/) and palˈp[e]bral (+ variant in [iː] / main pronunciation /100/).

92 Clearly, these cases are marginal and present a great deal of variation. Therefore, the surfacegeneralisation holds. Let us now consider what can be predicted. Let us first see if we canpredict the stress behaviour of derivatives based on properties of their bases. We can reliablydo so if they have a closed penult because of the strict permanence of consonants. However, asGuierre (1979: §2.5.9) points out, this strict permanence differs radically from the fundamentalvariability of vowels which can alternate between short and long or full and reduced. Burzio(1994: 54-55) illustrates this contrast by the asymmetry in the ability of vowels and consonantsto retain stress, e.g. asˈsist → asˈsistant vs. reˈv[ɪə]re → ˈrev[ə]rent.

93 Let us look at the vowel, in bases, which becomes the penult in derivatives and how the natureof that vowel can be linked with the stress pattern of derivatives. Let us assume that this vowelis underlying. In this part of the discussion, we purposefully ignore variants in /(-)1000/ forclarity (which are mainly found in neoclassical compounds). We leave out -ical derivativesas the vowel we are looking at is often absent in the base or is that of the suffix -ic, in whichcase stress never shifts except in the case of exceptionally stressed -ic words.36 We also treatseparately the words in which the vowel under consideration is an unstressed orthographic or <y> realized [ɪ], in which it is impossible to determine whether we are dealing with areduced vowel or a full short vowel. We found 15 such cases, which are detailed in (22).

(22) The vowel is orthographic or <y> realized [ɪ]:

a. the four remaining -ival words, all with penultimate stress: adjectival, gerundival, infinitival,substantival ©b. doctrinal /010/ ~ /100/ and pyramidal/0100/ ~ /1000/ ~ /2010/c. 9 cases with antepenultimate stress: aborigine, apocryphal, arbitral, digital, maximal, minimal,optimal, orbital, zenithal

94 The rest of the results are detailed in (23-25).

(23) The vowel is long (and may have a reduced variant):

a. 30 cases of stress shift to the penult: 9/10 words in -oidal, 12 in -cidal, anecdotal, archetypal,archival, hormonal, menopausal, ogival, paradisal, prodromal, ribosomal ©b. 14 cases of stress preservation on the penult: 7 words in-hedral, archducal, auroral, cerebral,cloacal, decretal, perineal ©c. 8 cases of variation between /(-)10/ and /(-)100/:communal, corneal, coronal, duodenal, ideal,intestinal, tracheal, vaginal ©d. 2 cases of /(-)100/: continual, residual ©

(24) The vowel is short (and may have a reduced variant): anˈtiphonal, ̍ cantonal (+ variant /010/),ˈsynodal and the 7 -iacal derived from words in -iac [-iæk] ©

(25) The vowel is reduceda. 4 cases of variation between /(-)10/ and /(-)100/:doctoral, electoral, mayoral, palatal, skeletal ©b. 85 cases of /(-)100/: all the derivatives in -{i, e, u}+VC0 in which the penultimate vowel ofthe derivative is present in the base, augural, clitoral, conjectural, ephemeral, femoral, natural,pastoral, personal, procedural, sculptural…

95 We summed up all those figures in Table 5.Table 5. Stress patterns of derivatives according to the nature of the penultimate vowel inthe base

Vowel in the baseStress patternof the derivative

<i, y> - [ɪ] V̅ V̆ Reduced V

/(-)10/ 4 44 0 0

/(-)10/ ~ /(-)100/ 2 8 1 4

/(-)100/ 9 2 9 85

96 These figures show a relatively solid tendency towards weight-based stress assignment in -al derivatives as long vowels do seem to “attract” stress when the suffix is added and shortor reduced vowels overall reject penultimate stress. The only cases that are problematic arethe four -ival cases in (20a), whose bases do not have long vowels in their final syllable. Toaccount for these cases, we could postulate underlying long vowels for these words only, whichwould be somehow ad hoc and would not offer any reason so as to why other words withreduced vowels should have underlying short vowels. A second option would be that adoptedby Trevian (2003), who adds a rule -val → /(-)10/, which is a rather odd generalisation. Whatis so special about /v/? And if the cause for this behaviour is not phonological, then what isthe status of -val?

97 However, even though the weight-based approach is satisfying for derivatives, it presentsissues for the rest of the corpus. We started by looking at derivatives, because it is the onlything we can do without postulating underlying vowels which are not attested elsewhere thanin the derivative whose pronunciation we are trying to predict. Doing so would constitute anattractive solution, as it would directly predict the correct surface patterns. The view accordingto which vowel values are lexical is widely shared and it predicts for example that underlyingschwas are unstressable. This seems rather strange because, if we see reduction as being aconsequence of stresslessness, then how can it also be its cause? Additionally, postulatingthat all vowels are lexical along with a few generalisations accounting for vowel shifts likethe Vowel Shift Rule, CiV Lengthening or Trisyllabic Shortening misses the fact that vowelvalues are widely conditioned by the environment in which vowels are found (see §2.2 andDabouis, forthcoming; Deschamps, 1994; Deschamps et al., 2004; Fournier 2010: Ch. 2 fora model of this conditioning).



98 In sum, the classic view consists in assigning stress based on what one postulates to bethe underlying vowels. This seems unsatisfactory because one can postulate any underlyingrepresentation to produce the correct surface form, which means this approach is non-predictive, and does not give a full account of vowel values. Therefore, the weight-basedapproach may offer a simpler solution with fewer generalisations even though it brings upwhat we regard as serious issues. The view developed in the Guierrian School, and in thispaper, seeks to predict both stress and vowel values and sees the latter as being, at least partly,conditioned by the presence or absence of the former. In that framework, an analysis of -almay include weight as we do find weight-related effects in words in -C2 + -al or in derivationfrom bases which contain long vowels, but syllable weight cannot satisfyingly account for allthe data.

99 Finally, if we consider all the entries of three syllables or more, excluding those in -C2+-al,-oidal and -ival, there are 658 entries out of 844 (78%) which have a proparoxytone stresspattern. If we consider that -ion is monosyllabic, then the 102 entries -ional can be added,which raises to total number of entries stressed on their antepenultimate syllable to 760 (90%).Therefore, a two-rule analysis of -al could be proposed with -C2+-al → /(-)10/ and VCal→ /(-)100/ but this would be a very simplified view of the empirical record. Such a simple system isa fairly good description of surface patterns and could very well be used for teaching purposes.However, when it comes to demonstration, we have shown the need for more parameters tobe taken into consideration.

8. Conclusion100 The situation that our study reveals is more complex than what most descriptions have

depicted. The stress patterns of -al adjectives seem to be regulated mostly by both the presenceof the suffix and certain elements preceding it (e.g. C2, suffixes like -iac and -oid or the root B inneoclassical compounds) and by larger generalisations, in which case the stress patterns cannotbe attributed to -al. Overall, derivational relationships exhibit direct computation of primarystress, but we have shown that it is not possible to demonstrate that -al alone is responsiblefor this behaviour, given that:

• the most regular sets can be described as formal micro-paradigms, i.e. they share formalproperties which may be held responsible for the regularisation of the set;

• the set of derived units which do not belong to any such micro-paradigms but to thelarger paradigm of adjectives in -al is the most irregular set.

101 Finally, we saw that a weight-based analysis of the stress behaviour of this suffix comes withsome heavy theoretical implications, namely the necessity of considering vowel values to belexical. In any case, this paper has shown the necessity of using empirical data because itmay allow us to discover that the empirical situation is far more heterogeneous than what theliterature suggests.

Bibliography

Anttila, Arto. “Morphologically Conditioned Phonological Alternations.” Natural Language &Linguistic Theory 20.1 (2002): 1–42.

Aronoff, Mark. Word Formation in Generative Grammar. Cambridge, MA: MIT Press, 1976.

Bauer, Laurie, Rochelle Lieber & Ingo Plag. The Oxford reference guide to English Morphology. Oxford:Oxford University Press, 2013.

Benua, Laura. Transderivational Identity: Phonological Relations between Words. Ph.D. dissertation,University of Massachussetts, 1997.

Bermúdez-Otero, Ricardo. “The Architecture of Grammar and the Division of Labour in Exponence.”The Morphology and Phonology of Exponence. Ed. Jochen Trommer, 2012. 8–83.

Bermúdez-Otero, Ricardo, and April McMahon. “English Phonology and Morphology.” The Handbookof English Linguistics. Ed. Jochen Trommer. Oxford: Oxford University Press, 2006. 382–410.

Burzio, Luigi. Principles of English Stress. Vol. 72. New York: Cambridge University Press, 1994.



Chomsky, Noam, and Morris Halle. The Sound Pattern of English. Cambridge, MA, London, England:MIT Press, 1968.

Collie, S. (2008). English Stress Preservation: the Case for “Fake Cyclicity.” English Language andLinguistics 12.3 (2008): 505–532.

Collie, Sarah. “English Stress Preservation and Stratal Optimality Theory.” Ph.D. dissertation, Universityof Edinburgh, 2007.

Dabouis, Quentin. “English Stress and Underlying Representations.” Proceedings of the firstPostgraduate and Academic Researchers in Linguistics at York 1 (2014): 1–15.

Dabouis, Quentin. English Phonology and the Literate Speaker English Word Stress: Theories, Data andVariation. Eds. Jacques Durand, Anne Przewozny, Eiji Yamada, forthcoming.

Dahak, Anissa. Etude diachronique, phonologique et morphologique des syllabes inaccentuées enanglais contemporain. Ph.D dissertation. Université de Paris Diderot, 2011.

Deschamps, Alain. De l’écrit à l'oral et de l'oral à l'écrit. Paris: Ophrys, 1994.

Deschamps, Alain, Jean-Louis Duchet, Jean-Michel Fournier & Michael O’Neil. English Phonology andGraphophonemics. Paris: Ophrys, 2004.

Domahs, Ulrike, Ingo Plag, and Rebecca Carroll. “Word Stress Assignment in German, English andDutch: Quantity-Sensitivity and Extrametricality Revisited.” The Journal of Comparative GermanicLinguistics 17.1 (2014): 59–96.

Fournier, Jean-Michel. “From a Latin Syllable-Driven Stress System to a Romance versus GermanicMorphology-Driven Dynamics: In Honour of Lionel Guierre.” Language Sciences 29 (2007): 218–236.

Fournier, Jean-Michel. Manuel D’anglais Oral. Paris: Ophrys, 2010.

Fournier, Jean-Michel. “Que Contraignent Les Terminaisons Contraignantes?” Topiques, NouvellesRecherches En Linguistique Anglaise, Travaux XCIII Du CIEREC. Université de Saint-Etienne, 1998.

Fournier, Pierre. “Accentuation et Prononciation Des Suffixés En -Ous En Anglais Contemporain.” Ph.D.dissertation, Université de Tours, 2011.

Giegerich, Heinz J. Lexical Strata in English: Morphological Causes, Phonological Effects. Cambridge:Cambridge University Press, 1999.

Girard, Isabelle. “Isomorphisme En Anglais Contemporain: Étude de Quelques Suffixes.” Ph.D.dissertation, Université de Tours, 2007.

Guierre, Lionel. Drills in English Stress Patterns: Ear and Speech Training Drills and Tests for Studentsof English as a Foreign Language. 3rd editio. Paris: Armand Colin - Longman, 1975.

Guierre, Lionel. “Essai Sur L’accentuation En Anglais Contemporain: Eléments Pour Une Synthèse.”Ph.D. dissertation, Université Paris-VII, 1979.

Harris, John & Edmund Gussmann. “Final codas: why the west was wrong.” Structure and Interpretation.Studies in Phonology. Ed. Eugeniuc Cyran. Lublin: Folium, 1998. 139–162.

Hay, Jennifer. “From Speech Perception to Morphology: Affix Ordering Revisited.” Linguistic Societyof America 78.3 (2002): 527–555.

Hayes, Bruce. A Metrical Theory of Stress Rules. Ph.D. dissertation. MIT, 1981.

Hayes, Bruce. “Extrametricality and English Stress.” Linguistic Inquiry 13.2 (1982): 227–276.

Hayes, Bruce. “The Phonology of Rhythm in English.” Linguistic Inquiry 15.1 (1984): 33–74.

Kaisse, Ellen M., and Patricia A. Shaw. “On the Theory of Lexical Phonology.” Phonology Yearbook2 (1985): 1–30.

Kiparsky, Paul. “From Cyclic Phonology to Lexical Phonology.” The Structure of PhonologicalRepresentations I. Ed. Harry Van der Hulst and Norval Smith. Dordrecht: Foris, 1982. 131–175.

Liberman, Mark, and Alan Prince. “On Stress and Linguistic Rhythm.” Linguistic inquiry 8.2 (1977):249–336.

Luick, Karl. “Beiträge Zur Englischen Grammatik III. Die Quantitäts Veränderungen Im Laufe DerEnglischen Sprachentwicklung.” Anglia 20 (1898): 335–362.

Marchand, Hans. “The Categories and Types of Present-Day English Word-Formation.” 1968.

Orgun, Cemil Orhan. “Sign-Based Morphology and Phonology: With Special Attention to OptimalityTheory.” Linguistics (1996).



Pater, Joe. “Non-Uniformity in English Secondary Stress: The Role of Ranked and Lexically SpecificConstraints.” Phonology 17 (2000): 237–274.

Pater, Joe. “On the Nonuniformity of Weight-to-Stress and Stress Preservation Effects in English.” Ms.,McGill University, 1995.

Raffelsiefen, Renate. “Phonological Restrictions on English Word-Formation.” HSK Word-Formation:An International Handbook of the Languages of Europe. Volume II. Ed. Peter O. Müller et al. Berlin,Boston: Mouton de Gruyter, 2015.

Raffelsiefen, Renate. “Relating Words: A Model of Base Recognition. Part I.” Linguistic Analysis 23(1993): 3–161.

Schane, Sanford A. “Rhythm, Accent, and Stress in English Words.” Linguistic Inquiry 10.3 (1979):483–502.

Selkirk, Elisabeth O. “The Role of Prosodic Categories in English Word Stress.” Linguistic inquiry 11.3(1980): 563–605.

Siegel, Dorothy C. Topics in English Morphology. Ph.D dissertation. MIT, 1974.

Tournier, Jean. Introduction descriptive à la lexicogénétique de l’anglais contemporain. Paris, Geneva:Champion - Slatkine, 1985.

Trevian, Ives. Morphoaccentologie et Processus D’affixation de L'anglais. Bern: Peter Lang, 2003.

Trocmé, Hélène. Aspects de la réduction des voyelles prétoniques en anglais moderne. Ph.D dissertation.Université Paris 7, 1975.

Wenszky, Nora. Secondary Stress in English Words. Budapest: Akademai Kiado, 2004.

DictionariesDavies, Mark. The Corpus of Contemporary American English: 425 million words, 1990-present. Online:http://corpus.byu.edu/coca

Jones, Daniel. Cambridge English Pronouncing Dictionary, 17th edition, edited by P. Roach, J. Setter& J. Hartman, Cambridge: CUP, 2006.

Oxford English Dictionary [Online] http://www.oed.com

Wells, John C. Longman Pronunciation Dictionary, 3rd edition, London: Longman, 2008.

Notes

1 I would like to thank Scott Dombrowski and two anonymous reviewers for their constructive comments.Responsibility for any mistake or omission is mine alone.2 Such a lack of empirical backup seems to be the rule rather than the exception. As Domahs et al.(2014) puts it, studies on English phonology “suffer from a scarcity of systematic empirical evidence”.Wenszky (2004: 12) also mentions that generalisations are often illustrated by “convenient samples”,which could be accepted if guarantees that the samples are overall representative of the full empiricalrecord were given, along with the description of that record. Unfortunately, such works are too scarcein the literature on English stress.3 At the time, Guierre computerised Daniel Jones’s pronouncing dictionary (35,000 words at the time),on which he based the work developed in his 1979 thesis.4 For discussion about the phonological role of these elements, see Dabouis (2014).5 Fournier (2010: 11-12) defines a lexical unit as semantically inseparable unit. He distinguishes twotypes of lexical units: those which are autonomous (words) and those which are not (prefixes). As aconsequence, the second principle may forbid stress clashes within lexical units, but it is quite commonto find such clashes in constructions made of two lexical units such as ˌreˈdo, ˌunˈfinished and ˌcoˈpilot.6 For certain suffixes, additional semantic information such as “agent” (e.g. -or in lessor, donor, vendor)can be taken into account.7 Angle brackets are used for orthography.8 That does not include foreign vowels (e.g. ban[ɑː]na, alb[iː]no, blas[eɪ]).9 See also a summarised version of these rules in Dabouis (forthcoming).10 Most Guierrian authors deal with stressed vowels only but Dahak (2011), Guierre (1979) and Trocmé(1975) also study unstressed vowels.11 These are my own translations from French, here are the original quotations: “Certaines séquencesfinales exercent une contrainte absolue en fixant immuablement l’accent primaire sur une syllabe

http://corpus.byu.edu/coca

http://www.oed.com/

donnée” (Trevian, 2003: 55). “Les suffixes contraignants imposent le calcul direct, c’est-à-dire lapertinence de la structure de l’unité et d’elle seule” (Fournier, 1998).12 This phenomenon is often called Trisyllabic Shortening or Laxing, but Guierrian authors usually talkabout “Luick’s rule” after Luick (1898), who first described the phenomenon.13 A reviewer points out the fact that -ic can be analysed as ending in an onset followed by an emptynucleus (Harris & Gussmann 1998). An alternative analysis consists in considering that suffixes aregenerally extrametrical, but that -ic is not (Hayes 1981). Both analyses can capture the facts but neitherproposes a reason why this suffix has these special properties and why other -VC suffixes such as -al,-ive or -ous do not.14 In the case of words in -{i, e, u}+V(C0(e)), the nature of the suffix involved (if there is one) is relevant.As Fournier (2010: 33) points out, if that suffix is a strictly neutral suffix, then the rule does not apply(e.g. deˈnial, ˈmarriage).15 We could add Selkirk (1980) to the list, even though these generalisations are captured by generalfeet well-formedness principles in her model.16 A surface-oriented approach could argue that the -val generalisation is not needed as they could fallinto the “preceding heavy syllable” category formed by the two preceding generalisations. However,Trevian does not do so, presumably because he assumes, as we do, that vowel length is computed afterthe position of stresses. See discussion in §7.17 Fournier’s (2010: 28) definition of functional consonant clusters (C2), which is adopted here, is thefollowing: A consonant cluster is defined as: <x>; Every cluster of at least two consonants, except <Ch>,<Cr> and <C + syllabic C>, and that includes consonant geminates.18 The initial study also included Australian English pronunciations from the Macquarie onlinedictionary and American English pronunciations from CEPD and LPD but, as minor variation was foundand for the sake of the clarity of the presentation, the study presented here will be concerned only withBritish English. Therefore, all the phonemic transcriptions below will be British English pronunciationsfrom CEPD or LPD.19 We are aware that some rare or specialised vocabulary will still be present in our corpus after that step.However, part of it (neoclassical compounds) will be treated separately as it has been shown previouslythat this part of the vocabulary is subject to different rules (Fournier, 2010: 75-78). We believe that thedifferences between different parts of the lexicon should be acknowledged, for not doing so might bias theresults. For example, the recurrent use of Amerindian names in the literature on English stress (e.g. eitherone or both of the two words Ticonderoga and Monongahela can be found in Burzio (1994), Chomsky& Halle (1968), Collie (2007), Hayes (1982, 1984), Kiparsky (1982), Liberman & Prince (1977), Pater(1995, 2000), Schane (1979), Selkirk (1980)) is never mentioned as such. We have no objection to theuse of these names, but it would seem reasonable to acknowledge that they are borrowings and that,therefore, they might not be representative of the global phonology of English, but only of a special setof rules or constraints applying to borrowings or even, more specifically, to Amerindian borrowings.Theoretically, this can be achieved through the use of “cophonologies” (Anttila 2002; Bermúdez-Otero &McMahon 2006; Collie 2007; Orgun 1996). We can add that they are also proper nouns and Raffelsiefen(1993: 90-93) claims that names and nouns are subject to different restrictions and, therefore, confusionbetween them should be avoided.20 Following Girard (2007), we refer to the double suffix -iacal and not -acal as can be found inFournier’s work. This is based on the fact that the only word in -acal not in -iacal is cloˈacal would bean exception to the rule -acal → /(-)100/.21 To be exact, this variable set does not include (Fournier, 2010: 101-143):words the first vowel of which is a digraph, which is always a V̅;words the first vowel of which is , which is always a V̅;words ending in -ic, which have ˈV̆;words ending in -

V#, which have ˈV̅;words ending in -iC#, which have ˈV̆.22 Full sets will not always be given when too large, but full sets will be marked with ©.23 As pointed out by a reviewer, these items can also be analysed as phonologically non-derived.24 This shows that Luick’s rule (see note Erreur : source de la référence non trouvée) does not apply.25 We consider -ion to be dissyllabic and therefore aspirational to have pre-antepenultimate stress. See§6.3 for discussion.26 These include non-derived units, for only derivation makes the test possible.

27 The set of entries in -inal is less regular than other sets chiefly because of irregular stress variants, butall present a regular variant: intestinal /0100/ ~ /2010/; doctrinal /010/ ~ /100/; matutinal /2010/ ~ /0100/;officinal /2010/ ~ /0100/; vaginal /010/ ~ /100/.28 This was translated from French: “si, dans une classe définie, un trait divise cette classe en deuxgroupes d’unités ayant un comportement distinct, et que ce trait est la seule différence entre ces deuxgroupes, alors c’est nécessairement lui qui est responsable de la différence de comportement.” (Fournier,1997: 60).29 Semi-attractive roots are roots that do not systematically retain stress.30 germicidal and suicidal present stress variants in /1000/.31 The phonemic transcriptions are from the OED.32 is not concerned by Luick’s rule as all cases of <ˈu> not followed by C2 are free vowels, e.g.ˈmusic, reˈjuvenate, ˈpurity.33 A reviewer questions the fact that artisanal, cerebral, coronal or integral are derived. The OEDdoes list artisan, cerebrum and integer which all share a semantic relationship with their correspondingderivatives. However, it is true that the last three examples refer to specialised vocabulary and that allspeakers may not be familiar with the bases. Consequently, the variation observed in the adjectives couldbe due to the fact that some speakers know their bases (and use the isomorphic variants) while others donot (and use direct computation, i.e. antepenultimate stress).34 A reviewer points out the fact that the situation might be different in American English, as is thecase for patronal, which has penultimate stress in British English but generally has antepenultimatestress in American English. As mentioned in note 18, our initial study included American and Australianpronunciations as well. We found very little variation, as only 27 words have variants across dictionariesor English varieties. Furthermore, sometimes that variation only consists in the reversal of the mainpronunciation and the variant, e.g. officinal: EPD BrE /2010/ ~ /0100/ vs. EPD AmE /0100/ ~ /2010/.This is consistent with Martin's (2011) finding that English stress is very stable across varieties, even inthe most unstable parts of the vocabulary.35 This is not surprising if we consider Dahak's (2011) that vowels tend to reduce in penults when primarystress is on the antepenult 96.6% of the time.36 For example, ˈheretic → heˈretical, politics → poˈlitical or aˈrithmetic → ˌarithˈmetical.

References

Electronic reference

Quentin Dabouis, « Is the Adjectival Suffix -al a Strong Suffix? », Anglophonia [Online],21 | 2016, Online since 01 July 2016, connection on 22 September 2016. URL : http://anglophonia.revues.org/754 ; DOI : 10.4000/anglophonia.754

Author

Quentin DabouisUniversité François-Rabelais, Tours - LLL (UMR 7270)[email protected]

Copyright

Anglophonia – French Journal of English Linguistics est mis à disposition selon les termesde la licence Creative Commons Attribution - Pas d'Utilisation Commerciale - Pas deModification 4.0 International.

Abstracts

Dans le cadre de l’approche introduite par Guierre (1979), cet article remet en question laclassification communément admise selon laquelle le suffixe adjectival -al serait un suffixecontraignant. Pour cela, nous avons eu recours à l’étude d’un large corpus constitué à partirde dictionnaires de prononciation et complété par d’autres sources telles que l’Oxford English

mailto:[email protected]

http://creativecommons.org/licenses/by-nc-nd/4.0/



Dictionary. Après avoir passé en revue les différentes analyses de ce suffixe dans la littérature,nous montrons que ces descriptions sont souvent inexactes. En effet, si l’on trouve bien -al dansplusieurs terminaisons contraignantes déclenchant un calcul direct du placement accentuel, cecomportement n’est observé que dans certains contextes spécifiques. Certaines sous-classesen -al se comportent de manière tout à fait régulière mais -al seul (c’est-à-dire lorsqu’il n’estpas associé à certains éléments spécifiques le précédant) ne déclenche pas le calcul direct defaçon systématique. Within the framework introduced by Guierre (1979), this paper challenges the commonlyaccepted classification of the adjectival suffix -al as a strong suffix through the study of acorpus of pronouncing dictionary data enriched with additional information from other sourcessuch as the Oxford English Dictionary. After a review of the disparities between differentanalyses and classifications of that suffix in the literature, it is shown that most descriptionspreviously proposed are problematic. We do find -al in strong endings triggering a directcomputation of stress placement, but only in certain specific subclasses. A number of setsending in -al are highly regular, but -al by itself (i.e. when it is not associated with certainelements preceding it) does not trigger direct computation systematically.

Index terms

Mots-clés : corpus, accentuation de l’anglais, phonologie, morphologie, suffixationKeywords : corpus study, English stress, phonology, morphology, suffixation

is the adjectival suffix -al a strong suffix?

Documents